https://lulietlyan.github.io Notes on computer networks, operating systems, MySQL, Redis, message queues, Golang, AI agents, and projects. en https://lulietlyan.github.io/topics/computer-network https://lulietlyan.github.io/topics/computer-network Sat, 17 Oct 2026 17:48:29 GMT TCP/IP、连接管理、可靠传输与 HTTP 应用层 https://lulietlyan.github.io/topics/computer-network/http-header-fields https://lulietlyan.github.io/topics/computer-network/http-header-fields Tue, 13 Oct 2026 00:00:00 GMT 梳理 HTTP 首部字段的常见用途，涵盖内容协商、缓存、跨源访问、反向代理与请求追踪。 https://lulietlyan.github.io/topics/operating-system https://lulietlyan.github.io/topics/operating-system Sat, 17 Oct 2026 17:48:29 GMT 进程与并发、虚拟内存、I/O 多路复用与死锁等操作系统笔记。 https://lulietlyan.github.io/topics/operating-system/select-poll-epoll https://lulietlyan.github.io/topics/operating-system/select-poll-epoll Wed, 09 Sep 2026 00:00:00 GMT 从数据结构、拷贝成本、就绪通知方式与规模扩展性，对比三种经典 I/O 多路复用接口。 https://lulietlyan.github.io/topics/operating-system/epoll-lt-et https://lulietlyan.github.io/topics/operating-system/epoll-lt-et Wed, 09 Sep 2026 00:00:00 GMT 说明 LT 与 ET 的通知条件、读写循环要求，以及为何 ET 常配合非阻塞 I/O。 https://lulietlyan.github.io/topics/operating-system/epoll-efficiency https://lulietlyan.github.io/topics/operating-system/epoll-efficiency Wed, 09 Sep 2026 00:00:00 GMT 从内核事件表、就绪队列和“只处理就绪集合”说明 epoll 相对 select/poll 的关键优势。 https://lulietlyan.github.io/topics/operating-system/malloc-brk-mmap https://lulietlyan.github.io/topics/operating-system/malloc-brk-mmap Wed, 09 Sep 2026 00:00:00 GMT 说明 glibc 等常见实现里小块走 brk/sbrk 堆扩展、大块常走 mmap，以及阈值策略差异。 https://lulietlyan.github.io/topics/operating-system/mmap https://lulietlyan.github.io/topics/operating-system/mmap Wed, 09 Sep 2026 00:00:00 GMT 解释内存映射把文件或匿名页映射进进程地址空间的机制，以及文件映射、匿名映射与共享的基本差别。 https://lulietlyan.github.io/topics/operating-system/memory-alignment https://lulietlyan.github.io/topics/operating-system/memory-alignment Wed, 09 Sep 2026 00:00:00 GMT 说明对齐的目的、自然对齐规则，以及它对结构体大小、性能和原子访问的影响。 https://lulietlyan.github.io/topics/operating-system/struct-sizeof-padding https://lulietlyan.github.io/topics/operating-system/struct-sizeof-padding Wed, 09 Sep 2026 00:00:00 GMT 按常见 32 位 ABI 逐步计算含 padding 的结构体大小，并解释为何结果是 20 字节。 https://lulietlyan.github.io/topics/operating-system/deadlock-conditions https://lulietlyan.github.io/topics/operating-system/deadlock-conditions Wed, 09 Sep 2026 00:00:00 GMT 说明死锁的四个必要条件：互斥、占有并等待、不可抢占、循环等待。 https://lulietlyan.github.io/topics/operating-system/deadlock-prevention https://lulietlyan.github.io/topics/operating-system/deadlock-prevention Wed, 09 Sep 2026 00:00:00 GMT 从破坏死锁四条件出发，说明预防、避免、检测恢复，以及工程上常用的加锁次序与超时策略。 https://lulietlyan.github.io/topics/SQL https://lulietlyan.github.io/topics/SQL Sat, 17 Oct 2026 17:48:29 GMT Notes on SQL, indexes, transactions and so on. https://lulietlyan.github.io/topics/SQL/MySQL https://lulietlyan.github.io/topics/SQL/MySQL Sat, 17 Oct 2026 17:48:29 GMT A source-guided deep dive into MySQL internals, covering storage engines, B+Tree indexes, transactions, locks, MVCC, concurrency control, query execution, and performance optimization. https://lulietlyan.github.io/topics/SQL/MySQL/mysql-redis-cache-consistency https://lulietlyan.github.io/topics/SQL/MySQL/mysql-redis-cache-consistency Fri, 28 Aug 2026 00:00:00 GMT 面试视角拆解 Cache Aside、并发脏写、删除失败补偿、延迟双删、Outbox 与 Binlog CDC，以及强一致性的边界。 https://lulietlyan.github.io/topics/redis https://lulietlyan.github.io/topics/redis Sat, 17 Oct 2026 17:48:29 GMT Notes on data structures, persistence, replication, and caching. https://lulietlyan.github.io/topics/redis/redis-data-structures https://lulietlyan.github.io/topics/redis/redis-data-structures Thu, 03 Sep 2026 00:00:00 GMT 区分 Redis 对外数据类型和内部编码，梳理 String、Hash、List、Set、ZSet、Stream 等结构的实现、复杂度与使用场景。 https://lulietlyan.github.io/topics/redis/redis-zset-skiplist https://lulietlyan.github.io/topics/redis/redis-zset-skiplist Thu, 03 Sep 2026 00:00:00 GMT 从节点层级、查找与插入路径、跨度排名、字典辅助索引和 Listpack 编码解释 Redis Sorted Set 的实现。 https://lulietlyan.github.io/topics/message-queue https://lulietlyan.github.io/topics/message-queue Sat, 17 Oct 2026 17:48:29 GMT Notes on async work, load shedding, and delivery semantics. https://lulietlyan.github.io/topics/golang https://lulietlyan.github.io/topics/golang Sat, 17 Oct 2026 17:48:29 GMT Notes on the language, concurrency, the standard library, and engineering. https://lulietlyan.github.io/topics/golang/go-garbage-collection https://lulietlyan.github.io/topics/golang/go-garbage-collection Thu, 03 Sep 2026 00:00:00 GMT 从三色标记、写屏障和并发标记清扫出发，解释 Go GC 的完整周期、触发条件、成本模型与调优方法。 https://lulietlyan.github.io/topics/golang/go-gc-fragmentation-and-large-objects https://lulietlyan.github.io/topics/golang/go-gc-fragmentation-and-large-objects Thu, 03 Sep 2026 00:00:00 GMT 区分内部碎片与外部碎片，解释 Go 的 Size Class、Span、页分配和大对象路径，以及大块内存申请失败时的工程处理。 https://lulietlyan.github.io/topics/golang/go-gmp-scheduler https://lulietlyan.github.io/topics/golang/go-gmp-scheduler Thu, 03 Sep 2026 00:00:00 GMT 解释 Goroutine、操作系统线程和逻辑处理器如何协作，以及运行队列、工作窃取、网络轮询、系统调用和抢占调度。 https://lulietlyan.github.io/topics/projects https://lulietlyan.github.io/topics/projects Sat, 17 Oct 2026 17:48:29 GMT 个人项目的开发记录，包括项目背景、功能设计、技术实现和更新情况。 https://lulietlyan.github.io/topics/projects/mini-ollama https://lulietlyan.github.io/topics/projects/mini-ollama Sat, 17 Oct 2026 17:48:29 GMT 基于 Go、Cobra、Gin 和 llama.cpp 的本地 GGUF 模型命令行工具。 https://lulietlyan.github.io/topics/projects/mini-ollama/01-environment https://lulietlyan.github.io/topics/projects/mini-ollama/01-environment Sat, 17 Oct 2026 17:48:29 GMT 构建 llama.cpp 的 llama-server、初始化 Go 项目并检查 CUDA 后端。 https://lulietlyan.github.io/topics/projects/mini-ollama/02-cmd https://lulietlyan.github.io/topics/projects/mini-ollama/02-cmd Sat, 17 Oct 2026 17:48:29 GMT 使用 Cobra 注册命令、解析参数，并启动本地模型服务。 https://lulietlyan.github.io/topics/projects/mini-ollama/03-service-api https://lulietlyan.github.io/topics/projects/mini-ollama/03-service-api Sat, 17 Oct 2026 17:48:29 GMT 说明 serve 模式的模型调度、聊天记录和 HTTP 接口。 https://lulietlyan.github.io/topics/projects/mini-ollama/04-models-and-import https://lulietlyan.github.io/topics/projects/mini-ollama/04-models-and-import Sat, 17 Oct 2026 17:48:29 GMT 说明 GGUF 模型目录、manifest 校验和 Hugging Face 导入流程。 https://lulietlyan.github.io/topics/projects/pseudoclaude https://lulietlyan.github.io/topics/projects/pseudoclaude Sat, 17 Oct 2026 17:48:29 GMT A source-guided series on PseudoClaude's terminal runtime, model and tool ecosystem, context lifecycle, security boundaries, and multi-agent collaboration. https://lulietlyan.github.io/topics/projects/pseudoclaude/01-project-architecture https://lulietlyan.github.io/topics/projects/pseudoclaude/01-project-architecture Fri, 21 Aug 2026 00:00:00 GMT How PseudoClaude assembles its TUI, runner, tools, state services, and security modules from the main process entry point. https://lulietlyan.github.io/topics/projects/pseudoclaude/02-tui-state-machine https://lulietlyan.github.io/topics/projects/pseudoclaude/02-tui-state-machine Fri, 21 Aug 2026 00:00:00 GMT How PseudoClaude turns user input, runner events, streaming output, and permission approvals into a Bubble Tea terminal state machine. https://lulietlyan.github.io/topics/projects/pseudoclaude/03-react-agent-loop https://lulietlyan.github.io/topics/projects/pseudoclaude/03-react-agent-loop Fri, 21 Aug 2026 00:00:00 GMT How the PseudoClaude runner closes the ReAct loop across model streaming, tool calls, result feedback, and stop conditions. https://lulietlyan.github.io/topics/projects/pseudoclaude/04-tool-system https://lulietlyan.github.io/topics/projects/pseudoclaude/04-tool-system Fri, 21 Aug 2026 00:00:00 GMT How PseudoClaude unifies tool definitions, registration, visibility filtering, execution routing, and structured error results. https://lulietlyan.github.io/topics/projects/pseudoclaude/05-context-and-memory https://lulietlyan.github.io/topics/projects/pseudoclaude/05-context-and-memory Fri, 21 Aug 2026 00:00:00 GMT How Conversation, Session, Compact, and Memory divide responsibility for PseudoClaude's short- and long-lived context state. https://lulietlyan.github.io/topics/projects/pseudoclaude/06-plan-mode-security https://lulietlyan.github.io/topics/projects/pseudoclaude/06-plan-mode-security Fri, 21 Aug 2026 00:00:00 GMT How PseudoClaude combines prompts, tool filtering, runtime validation, and permission rules to constrain side effects in Plan Mode. https://lulietlyan.github.io/topics/projects/pseudoclaude/07-provider-stream-events https://lulietlyan.github.io/topics/projects/pseudoclaude/07-provider-stream-events Sun, 30 Aug 2026 00:00:00 GMT How PseudoClaude normalizes messages, tool calls, streaming, usage, cancellation, and errors across Anthropic and OpenAI providers. https://lulietlyan.github.io/topics/projects/pseudoclaude/08-mcp-tool-bridge https://lulietlyan.github.io/topics/projects/pseudoclaude/08-mcp-tool-bridge Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude turns a configured MCP server into a model-visible tool and routes each call back to the remote session. https://lulietlyan.github.io/topics/projects/pseudoclaude/09-skill-progressive-disclosure https://lulietlyan.github.io/topics/projects/pseudoclaude/09-skill-progressive-disclosure Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude exposes compact skill metadata first, then loads full SOP content and specialized tools only when requested. https://lulietlyan.github.io/topics/projects/pseudoclaude/10-session-jsonl-recovery https://lulietlyan.github.io/topics/projects/pseudoclaude/10-session-jsonl-recovery Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude persists messages and replacement snapshots in JSONL and replays them into a valid conversation after restart. https://lulietlyan.github.io/topics/projects/pseudoclaude/11-layered-instructions https://lulietlyan.github.io/topics/projects/pseudoclaude/11-layered-instructions Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude loads layered PSEUDOCLAUDE.md files, expands relative references safely, and builds a stable system prompt. https://lulietlyan.github.io/topics/projects/pseudoclaude/12-async-long-term-memory https://lulietlyan.github.io/topics/projects/pseudoclaude/12-async-long-term-memory Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude asynchronously extracts structured memory operations and updates project and user stores without blocking the active turn. https://lulietlyan.github.io/topics/projects/pseudoclaude/13-two-stage-context-compaction https://lulietlyan.github.io/topics/projects/pseudoclaude/13-two-stage-context-compaction Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude spills large tool results to disk and summarizes older history near the context limit while preserving recent tool-call closure. https://lulietlyan.github.io/topics/projects/pseudoclaude/14-dangerous-command-guard https://lulietlyan.github.io/topics/projects/pseudoclaude/14-dangerous-command-guard Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude normalizes run_command arguments and hard-denies known destructive patterns before configurable rules and permission modes. https://lulietlyan.github.io/topics/projects/pseudoclaude/15-workspace-path-sandbox https://lulietlyan.github.io/topics/projects/pseudoclaude/15-workspace-path-sandbox Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude resolves file targets, symlinks, and not-yet-created paths before rejecting access outside the active workspace. https://lulietlyan.github.io/topics/projects/pseudoclaude/16-layered-permission-rules https://lulietlyan.github.io/topics/projects/pseudoclaude/16-layered-permission-rules Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude evaluates session, local, project, and user permission rules before applying mode fallbacks and reusable approvals. https://lulietlyan.github.io/topics/projects/pseudoclaude/17-interactive-approval-flow https://lulietlyan.github.io/topics/projects/pseudoclaude/17-interactive-approval-flow Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude turns an Ask decision into a blocking Agent-to-TUI approval round trip and returns scoped authorization to tool execution. https://lulietlyan.github.io/topics/projects/pseudoclaude/18-plan-mode-double-guard https://lulietlyan.github.io/topics/projects/pseudoclaude/18-plan-mode-double-guard Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude constrains Plan Mode through prompts, read-only tool definitions, and a second runtime safety check before execution. https://lulietlyan.github.io/topics/projects/pseudoclaude/19-agent-delegation-models https://lulietlyan.github.io/topics/projects/pseudoclaude/19-agent-delegation-models Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude routes Agent tool calls into conversation forks or predefined roles with scoped models, tools, permissions, and execution contexts. https://lulietlyan.github.io/topics/projects/pseudoclaude/20-background-task-lifecycle https://lulietlyan.github.io/topics/projects/pseudoclaude/20-background-task-lifecycle Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude runs child agents in the foreground or as detached background tasks with status snapshots, cancellation, continuation, and TUI notifications. https://lulietlyan.github.io/topics/projects/pseudoclaude/21-team-lead-collaboration https://lulietlyan.github.io/topics/projects/pseudoclaude/21-team-lead-collaboration Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude models persistent teams, launches isolated in-process members, shares dependency-aware tasks, and reports member completion to the Lead. https://lulietlyan.github.io/topics/projects/pseudoclaude/22-mailbox-communication https://lulietlyan.github.io/topics/projects/pseudoclaude/22-mailbox-communication Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude persists per-agent mailboxes, serializes concurrent writers, injects unread messages as reminders, and wakes idle team members. https://lulietlyan.github.io/topics/projects/pseudoclaude/23-git-worktree-isolation https://lulietlyan.github.io/topics/projects/pseudoclaude/23-git-worktree-isolation Fri, 28 Aug 2026 00:00:00 GMT How PseudoClaude gives agents independent Git worktrees and branches, redirects execution through per-run CWDs, and protects local changes during cleanup. https://lulietlyan.github.io/topics/programming-languages https://lulietlyan.github.io/topics/programming-languages Sat, 17 Oct 2026 17:48:29 GMT 编程语言的类型系统、执行模型、运行时、并发模型与跨语言设计取舍。 https://lulietlyan.github.io/topics/programming-languages/java-python-go-comparison https://lulietlyan.github.io/topics/programming-languages/java-python-go-comparison Thu, 03 Sep 2026 00:00:00 GMT 从类型系统、执行方式、并发模型、内存管理、错误处理、部署和适用场景系统比较 Java、Python 与 Go。 https://lulietlyan.github.io/topics/programming-languages/compiled-vs-interpreted https://lulietlyan.github.io/topics/programming-languages/compiled-vs-interpreted Thu, 03 Sep 2026 00:00:00 GMT 从编译时机、目标代码、解释器、字节码、JIT 和 AOT 说明现代语言执行模型及常见误区。 https://lulietlyan.github.io/topics/programming-languages/malloc-vs-new https://lulietlyan.github.io/topics/programming-languages/malloc-vs-new Wed, 09 Sep 2026 00:00:00 GMT 从语言归属、构造析构、失败语义、可重载性与配对释放方式对比 C 的 malloc 与 C++ 的 new。 https://lulietlyan.github.io/topics/ai-agent https://lulietlyan.github.io/topics/ai-agent Sat, 17 Oct 2026 17:48:29 GMT LLM、VLM、RLHF、Agent、RAG 与模型评估面试问答，以及 AI 编程、Skill、MCP 和《深入理解 AI Agent》读书笔记。 https://lulietlyan.github.io/topics/ai-agent/ai-coding-tools-and-views https://lulietlyan.github.io/topics/ai-agent/ai-coding-tools-and-views Wed, 09 Sep 2026 00:00:00 GMT 从辅助边界说明常见 AI 编程工具的定位，以及审查、测试与责任仍在开发者一侧。 https://lulietlyan.github.io/topics/ai-agent/skill-agent-mcp https://lulietlyan.github.io/topics/ai-agent/skill-agent-mcp Wed, 09 Sep 2026 00:00:00 GMT 用标准术语区分 Agent 循环、Skill 工作方法与 MCP 工具协议，并说明三者如何协作。 https://lulietlyan.github.io/topics/ai-agent/llm https://lulietlyan.github.io/topics/ai-agent/llm Sat, 17 Oct 2026 17:48:29 GMT Transformer、位置编码、解码策略、词元化、MoE 与大规模训练。 https://lulietlyan.github.io/topics/ai-agent/llm/self-attention https://lulietlyan.github.io/topics/ai-agent/llm/self-attention Sat, 12 Sep 2026 00:00:00 GMT 从 Q、K、V 到缩放点积注意力，解释并行计算与长距离依赖。 https://lulietlyan.github.io/topics/ai-agent/llm/positional-encoding https://lulietlyan.github.io/topics/ai-agent/llm/positional-encoding Sat, 12 Sep 2026 00:00:00 GMT 介绍位置信息的作用，对比正弦余弦编码与可学习的绝对位置编码。 https://lulietlyan.github.io/topics/ai-agent/llm/rotary-position-embedding https://lulietlyan.github.io/topics/ai-agent/llm/rotary-position-embedding Sat, 12 Sep 2026 00:00:00 GMT 整理旋转位置编码的工作原理、相对位置特性与优劣势。 https://lulietlyan.github.io/topics/ai-agent/llm/mha-mqa-gqa https://lulietlyan.github.io/topics/ai-agent/llm/mha-mqa-gqa Sat, 12 Sep 2026 00:00:00 GMT 比较注意力头的组织方式，以及模型效果、KV Cache 和推理成本的权衡。 https://lulietlyan.github.io/topics/ai-agent/llm/transformer-architectures https://lulietlyan.github.io/topics/ai-agent/llm/transformer-architectures Sat, 12 Sep 2026 00:00:00 GMT 按结构、注意力机制和任务类型比较三种 Transformer 架构。 https://lulietlyan.github.io/topics/ai-agent/llm/scaling-laws https://lulietlyan.github.io/topics/ai-agent/llm/scaling-laws Sat, 12 Sep 2026 00:00:00 GMT 解释尺度定律及其对参数规模、数据规模和训练预算分配的指导意义。 https://lulietlyan.github.io/topics/ai-agent/llm/decoding-strategies https://lulietlyan.github.io/topics/ai-agent/llm/decoding-strategies Sat, 12 Sep 2026 00:00:00 GMT 比较贪心搜索、束搜索、Top-K 与 Top-P 采样的原理和优缺点。 https://lulietlyan.github.io/topics/ai-agent/llm/bpe-wordpiece https://lulietlyan.github.io/topics/ai-agent/llm/bpe-wordpiece Sat, 12 Sep 2026 00:00:00 GMT 从子词切分解释词元化，并比较 BPE 与 WordPiece 的合并规则。 https://lulietlyan.github.io/topics/ai-agent/llm/nlp-and-llm https://lulietlyan.github.io/topics/ai-agent/llm/nlp-and-llm Sat, 12 Sep 2026 00:00:00 GMT 从研究范围、技术范式、数据需求与应用方式理解 NLP 和 LLM 的关系。 https://lulietlyan.github.io/topics/ai-agent/llm/l1-l2-regularization https://lulietlyan.github.io/topics/ai-agent/llm/l1-l2-regularization Sat, 12 Sep 2026 00:00:00 GMT 解释两种正则化的惩罚项，以及稀疏化、特征选择和泛化能力的权衡。 https://lulietlyan.github.io/topics/ai-agent/llm/emergent-abilities https://lulietlyan.github.io/topics/ai-agent/llm/emergent-abilities Sat, 12 Sep 2026 00:00:00 GMT 整理涌现能力的含义、典型表现及原材料对模型规模的讨论。 https://lulietlyan.github.io/topics/ai-agent/llm/activation-functions https://lulietlyan.github.io/topics/ai-agent/llm/activation-functions Sat, 12 Sep 2026 00:00:00 GMT 介绍非线性的作用，以及 GeLU、SwiGLU 的原理和使用动机。 https://lulietlyan.github.io/topics/ai-agent/llm/mixture-of-experts https://lulietlyan.github.io/topics/ai-agent/llm/mixture-of-experts Sat, 12 Sep 2026 00:00:00 GMT 介绍专家网络、路由器、Top-K 稀疏激活与加权输出。 https://lulietlyan.github.io/topics/ai-agent/llm/large-scale-training https://lulietlyan.github.io/topics/ai-agent/llm/large-scale-training Sat, 12 Sep 2026 00:00:00 GMT 整理显存、分布式通信、训练稳定性等挑战及相应的工程手段。 https://lulietlyan.github.io/topics/ai-agent/llm/open-source-frameworks https://lulietlyan.github.io/topics/ai-agent/llm/open-source-frameworks Sat, 12 Sep 2026 00:00:00 GMT 整理训练、模型生态与推理框架，并保留 Qwen、DeepSeek 的论文阅读提示。 https://lulietlyan.github.io/topics/ai-agent/llm/frontier-paper-discussion https://lulietlyan.github.io/topics/ai-agent/llm/frontier-paper-discussion Sat, 12 Sep 2026 00:00:00 GMT 保留原材料的开放题提示，围绕问题、方法和对比实验组织论文介绍。 https://lulietlyan.github.io/topics/ai-agent/vlm https://lulietlyan.github.io/topics/ai-agent/vlm Sat, 17 Oct 2026 17:48:29 GMT 视觉语言对齐、CLIP、视觉指令微调、视频理解与多模态应用。 https://lulietlyan.github.io/topics/ai-agent/vlm/multimodal-alignment https://lulietlyan.github.io/topics/ai-agent/vlm/multimodal-alignment Sat, 12 Sep 2026 00:00:00 GMT 从模态鸿沟出发，整理多模态表示对齐与信息融合的主要思路。 https://lulietlyan.github.io/topics/ai-agent/vlm/clip-contrastive-learning https://lulietlyan.github.io/topics/ai-agent/vlm/clip-contrastive-learning Sat, 12 Sep 2026 00:00:00 GMT 介绍双编码器、图文配对、对比学习目标与零样本应用。 https://lulietlyan.github.io/topics/ai-agent/vlm/vision-language-connectors https://lulietlyan.github.io/topics/ai-agent/vlm/vision-language-connectors Sat, 12 Sep 2026 00:00:00 GMT 介绍视觉编码器、连接器和语言模型之间的架构关系与训练过程。 https://lulietlyan.github.io/topics/ai-agent/vlm/visual-instruction-tuning https://lulietlyan.github.io/topics/ai-agent/vlm/visual-instruction-tuning Sat, 12 Sep 2026 00:00:00 GMT 解释视觉指令数据、训练目标及其对对话和指令遵循能力的作用。 https://lulietlyan.github.io/topics/ai-agent/vlm/video-understanding https://lulietlyan.github.io/topics/ai-agent/vlm/video-understanding Sat, 12 Sep 2026 00:00:00 GMT 整理时序建模、帧间信息、计算开销与长视频理解的挑战。 https://lulietlyan.github.io/topics/ai-agent/vlm/visual-grounding https://lulietlyan.github.io/topics/ai-agent/vlm/visual-grounding Sat, 12 Sep 2026 00:00:00 GMT 介绍文本与图像区域的对应关系，以及定位任务和评估指标。 https://lulietlyan.github.io/topics/ai-agent/vlm/vlm-architectures https://lulietlyan.github.io/topics/ai-agent/vlm/vlm-architectures Sat, 12 Sep 2026 00:00:00 GMT 按原材料对比连接器与跨模态注意力范式的结构、优势和代价。 https://lulietlyan.github.io/topics/ai-agent/vlm/high-resolution-images https://lulietlyan.github.io/topics/ai-agent/vlm/high-resolution-images Sat, 12 Sep 2026 00:00:00 GMT 整理图像切块、动态分辨率等思路及视觉细节与计算量之间的权衡。 https://lulietlyan.github.io/topics/ai-agent/vlm/visual-hallucination https://lulietlyan.github.io/topics/ai-agent/vlm/visual-hallucination Sat, 12 Sep 2026 00:00:00 GMT 结合对象、属性与关系等例子，说明多模态生成中的幻觉表现。 https://lulietlyan.github.io/topics/ai-agent/vlm/vlm-applications https://lulietlyan.github.io/topics/ai-agent/vlm/vlm-applications Sat, 12 Sep 2026 00:00:00 GMT 整理多模态模型在具身智能、文档理解及其他场景中的应用潜力。 https://lulietlyan.github.io/topics/ai-agent/vlm/vlm-finetuning-experience https://lulietlyan.github.io/topics/ai-agent/vlm/vlm-finetuning-experience Sat, 12 Sep 2026 00:00:00 GMT 以原材料中的工业质检回答范例，梳理模型选择、数据准备、微调与评估。 https://lulietlyan.github.io/topics/ai-agent/rlhf https://lulietlyan.github.io/topics/ai-agent/rlhf Sat, 17 Oct 2026 17:48:29 GMT SFT、奖励模型、PPO、DPO、GRPO、信用分配与 AI 反馈对齐。 https://lulietlyan.github.io/topics/ai-agent/rlhf/rlhf-versus-sft https://lulietlyan.github.io/topics/ai-agent/rlhf/rlhf-versus-sft Sat, 12 Sep 2026 00:00:00 GMT 解释模仿示范与学习偏好的差异，以及 RLHF 希望解决的问题。 https://lulietlyan.github.io/topics/ai-agent/rlhf/rlhf-pipeline https://lulietlyan.github.io/topics/ai-agent/rlhf/rlhf-pipeline Sat, 12 Sep 2026 00:00:00 GMT 按输入、输出和目标梳理监督微调、奖励模型训练与策略优化。 https://lulietlyan.github.io/topics/ai-agent/rlhf/pairwise-preferences https://lulietlyan.github.io/topics/ai-agent/rlhf/pairwise-preferences Sat, 12 Sep 2026 00:00:00 GMT 比较相对偏好与绝对评分的标注优势、信息损失和偏差来源。 https://lulietlyan.github.io/topics/ai-agent/rlhf/reward-model https://lulietlyan.github.io/topics/ai-agent/rlhf/reward-model Sat, 12 Sep 2026 00:00:00 GMT 介绍模型架构、与策略模型的关系及 Bradley-Terry 偏好损失。 https://lulietlyan.github.io/topics/ai-agent/rlhf/ppo-and-kl-penalty https://lulietlyan.github.io/topics/ai-agent/rlhf/ppo-and-kl-penalty Sat, 12 Sep 2026 00:00:00 GMT 整理 PPO 的稳定性考虑、与其他算法的比较及 KL 约束的作用。 https://lulietlyan.github.io/topics/ai-agent/rlhf/kl-coefficient-tuning https://lulietlyan.github.io/topics/ai-agent/rlhf/kl-coefficient-tuning Sat, 12 Sep 2026 00:00:00 GMT 讨论对齐收益与策略偏移的平衡，以及监控指标和系数调整方法。 https://lulietlyan.github.io/topics/ai-agent/rlhf/reward-hacking https://lulietlyan.github.io/topics/ai-agent/rlhf/reward-hacking Sat, 12 Sep 2026 00:00:00 GMT 通过应用场景说明奖励代理被利用的问题，并整理缓解策略。 https://lulietlyan.github.io/topics/ai-agent/rlhf/direct-preference-optimization https://lulietlyan.github.io/topics/ai-agent/rlhf/direct-preference-optimization Sat, 12 Sep 2026 00:00:00 GMT 比较直接偏好优化与传统 RLHF 的流程、组件、训练复杂度和成本。 https://lulietlyan.github.io/topics/ai-agent/rlhf/alignment-failure-analysis https://lulietlyan.github.io/topics/ai-agent/rlhf/alignment-failure-analysis Sat, 12 Sep 2026 00:00:00 GMT 从奖励模型、偏好数据与训练过程分析对齐失效及改进方向。 https://lulietlyan.github.io/topics/ai-agent/rlhf/grpo-versus-ppo https://lulietlyan.github.io/topics/ai-agent/rlhf/grpo-versus-ppo Sat, 12 Sep 2026 00:00:00 GMT 保留原材料关于 GRPO 的论文阅读提示；原文未提供具体比较答案。 https://lulietlyan.github.io/topics/ai-agent/rlhf/gspo-dapo-grpo https://lulietlyan.github.io/topics/ai-agent/rlhf/gspo-dapo-grpo Sat, 12 Sep 2026 00:00:00 GMT 保留原材料的前沿算法阅读提示；原文未展开三种算法的比较。 https://lulietlyan.github.io/topics/ai-agent/rlhf/credit-assignment https://lulietlyan.github.io/topics/ai-agent/rlhf/credit-assignment Sat, 12 Sep 2026 00:00:00 GMT 整理奖励归因问题，以及不同奖励粒度对训练信号的影响。 https://lulietlyan.github.io/topics/ai-agent/rlhf/rlaif https://lulietlyan.github.io/topics/ai-agent/rlhf/rlaif Sat, 12 Sep 2026 00:00:00 GMT 介绍使用 AI 反馈进行对齐的流程、扩展性与潜在偏差。 https://lulietlyan.github.io/topics/ai-agent/agent https://lulietlyan.github.io/topics/ai-agent/agent Sat, 17 Oct 2026 17:48:29 GMT Agent 组件、ReAct、规划、记忆、工具调用、多智能体协作与微调。 https://lulietlyan.github.io/topics/ai-agent/agent/agent-components https://lulietlyan.github.io/topics/ai-agent/agent/agent-components Sat, 12 Sep 2026 00:00:00 GMT 从感知、规划、记忆与工具使用等模块理解智能体的运行方式。 https://lulietlyan.github.io/topics/ai-agent/agent/react https://lulietlyan.github.io/topics/ai-agent/agent/react Sat, 12 Sep 2026 00:00:00 GMT 通过思考、行动、观察的循环介绍 ReAct 框架及其工作流程。 https://lulietlyan.github.io/topics/ai-agent/agent/agent-planning https://lulietlyan.github.io/topics/ai-agent/agent/agent-planning Sat, 12 Sep 2026 00:00:00 GMT 整理 CoT、ToT、GoT 等规划思路及任务分解方法。 https://lulietlyan.github.io/topics/ai-agent/agent/agent-memory https://lulietlyan.github.io/topics/ai-agent/agent/agent-memory Sat, 12 Sep 2026 00:00:00 GMT 比较上下文、摘要、向量数据库等记忆机制及其用途。 https://lulietlyan.github.io/topics/ai-agent/agent/tool-use-function-calling https://lulietlyan.github.io/topics/ai-agent/agent/tool-use-function-calling Sat, 12 Sep 2026 00:00:00 GMT 从 Function Calling 解释工具描述、参数生成、调用执行与结果反馈。 https://lulietlyan.github.io/topics/ai-agent/agent/langchain-versus-llamaindex https://lulietlyan.github.io/topics/ai-agent/agent/langchain-versus-llamaindex Sat, 12 Sep 2026 00:00:00 GMT 按原材料比较逻辑编排与数据检索两类开发需求及框架特点。 https://lulietlyan.github.io/topics/ai-agent/agent/agent-engineering-challenges https://lulietlyan.github.io/topics/ai-agent/agent/agent-engineering-challenges Sat, 12 Sep 2026 00:00:00 GMT 整理规划可靠性、工具调用、上下文管理、评估和安全等工程问题。 https://lulietlyan.github.io/topics/ai-agent/agent/multi-agent-systems https://lulietlyan.github.io/topics/ai-agent/agent/multi-agent-systems Sat, 12 Sep 2026 00:00:00 GMT 比较单 Agent 与多 Agent 的分工、协作、通信成本和协调难题。 https://lulietlyan.github.io/topics/ai-agent/agent/embodied-agents https://lulietlyan.github.io/topics/ai-agent/agent/embodied-agents Sat, 12 Sep 2026 00:00:00 GMT 从感知、行动空间、环境动态与反馈解释机器人和游戏中的智能体。 https://lulietlyan.github.io/topics/ai-agent/agent/agent-safety-alignment https://lulietlyan.github.io/topics/ai-agent/agent/agent-safety-alignment Sat, 12 Sep 2026 00:00:00 GMT 整理模型对齐、权限约束、人类监督、环境隔离与审计机制。 https://lulietlyan.github.io/topics/ai-agent/agent/a2a-protocol https://lulietlyan.github.io/topics/ai-agent/agent/a2a-protocol Sat, 12 Sep 2026 00:00:00 GMT 从个体实现与跨 Agent 交互标准的侧重点解释 A2A。 https://lulietlyan.github.io/topics/ai-agent/agent/agent-framework-selection https://lulietlyan.github.io/topics/ai-agent/agent/agent-framework-selection Sat, 12 Sep 2026 00:00:00 GMT 保留选型回答范例，围绕任务成功率、过程效率和鲁棒性组织回答。 https://lulietlyan.github.io/topics/ai-agent/agent/agent-finetuning-data https://lulietlyan.github.io/topics/ai-agent/agent/agent-finetuning-data Sat, 12 Sep 2026 00:00:00 GMT 整理决策轨迹与教师模型合成数据，并保留人工和真实交互数据的提纲。 https://lulietlyan.github.io/topics/ai-agent/rag https://lulietlyan.github.io/topics/ai-agent/rag Sat, 17 Oct 2026 17:48:29 GMT 知识库构建、切块、嵌入、检索优化、知识图谱、评估与部署。 https://lulietlyan.github.io/topics/ai-agent/rag/rag-versus-finetuning https://lulietlyan.github.io/topics/ai-agent/rag/rag-versus-finetuning Sat, 12 Sep 2026 00:00:00 GMT 介绍检索增强生成的基本流程，以及知识更新、可追溯性与成本优势。 https://lulietlyan.github.io/topics/ai-agent/rag/rag-pipeline https://lulietlyan.github.io/topics/ai-agent/rag/rag-pipeline Sat, 12 Sep 2026 00:00:00 GMT 从离线数据准备、索引构建到在线检索、提示组装和生成逐步梳理。 https://lulietlyan.github.io/topics/ai-agent/rag/chunking-strategies https://lulietlyan.github.io/topics/ai-agent/rag/chunking-strategies Sat, 12 Sep 2026 00:00:00 GMT 讨论语义完整性、检索精度、上下文容量与索引成本的权衡。 https://lulietlyan.github.io/topics/ai-agent/rag/embedding-model-selection https://lulietlyan.github.io/topics/ai-agent/rag/embedding-model-selection Sat, 12 Sep 2026 00:00:00 GMT 从任务匹配、语言领域、检索指标和部署开销整理嵌入模型选型。 https://lulietlyan.github.io/topics/ai-agent/rag/retrieval-improvements https://lulietlyan.github.io/topics/ai-agent/rag/retrieval-improvements Sat, 12 Sep 2026 00:00:00 GMT 整理混合检索、重排序、查询改写及其他检索优化手段。 https://lulietlyan.github.io/topics/ai-agent/rag/lost-in-the-middle https://lulietlyan.github.io/topics/ai-agent/rag/lost-in-the-middle Sat, 12 Sep 2026 00:00:00 GMT 解释长上下文中间信息利用不足的现象及 RAG 中的应对方法。 https://lulietlyan.github.io/topics/ai-agent/rag/rag-evaluation https://lulietlyan.github.io/topics/ai-agent/rag/rag-evaluation Sat, 12 Sep 2026 00:00:00 GMT 分别整理检索质量、生成质量与系统整体表现的评估指标。 https://lulietlyan.github.io/topics/ai-agent/rag/knowledge-graph-retrieval https://lulietlyan.github.io/topics/ai-agent/rag/knowledge-graph-retrieval Sat, 12 Sep 2026 00:00:00 GMT 围绕结构化关系、多跳推理和可解释查询讨论图检索的适用场景。 https://lulietlyan.github.io/topics/ai-agent/rag/adaptive-iterative-rag https://lulietlyan.github.io/topics/ai-agent/rag/adaptive-iterative-rag Sat, 12 Sep 2026 00:00:00 GMT 介绍迭代检索、自适应检索等超出单次检索生成流程的范式。 https://lulietlyan.github.io/topics/ai-agent/rag/rag-deployment-challenges https://lulietlyan.github.io/topics/ai-agent/rag/rag-deployment-challenges Sat, 12 Sep 2026 00:00:00 GMT 整理数据质量、知识更新、检索效果、延迟成本与运维问题。 https://lulietlyan.github.io/topics/ai-agent/rag/search-versus-rag https://lulietlyan.github.io/topics/ai-agent/rag/search-versus-rag Sat, 12 Sep 2026 00:00:00 GMT 从目标、输出、系统流程和评估方式比较搜索与检索增强生成。 https://lulietlyan.github.io/topics/ai-agent/rag/rag-framework-selection https://lulietlyan.github.io/topics/ai-agent/rag/rag-framework-selection Sat, 12 Sep 2026 00:00:00 GMT 按原材料整理 RAGFlow、Haystack、DSPy 等工具及场景选择思路。 https://lulietlyan.github.io/topics/ai-agent/evaluation https://lulietlyan.github.io/topics/ai-agent/evaluation Sat, 17 Oct 2026 17:48:29 GMT LLM 基准、模型裁判、Agent 过程指标、红队测试、人工评估与线上监控。 https://lulietlyan.github.io/topics/ai-agent/evaluation/bleu-rouge-limitations https://lulietlyan.github.io/topics/ai-agent/evaluation/bleu-rouge-limitations Sat, 12 Sep 2026 00:00:00 GMT 讨论词面重合指标在语义、开放生成、事实性和推理评估上的局限。 https://lulietlyan.github.io/topics/ai-agent/evaluation/llm-benchmarks https://lulietlyan.github.io/topics/ai-agent/evaluation/llm-benchmarks Sat, 12 Sep 2026 00:00:00 GMT 介绍 MMLU、Big-Bench、HumanEval 等基准及其评估重点。 https://lulietlyan.github.io/topics/ai-agent/evaluation/llm-as-a-judge https://lulietlyan.github.io/topics/ai-agent/evaluation/llm-as-a-judge Sat, 12 Sep 2026 00:00:00 GMT 整理模型裁判的使用方式、规模化优势及位置、长度和自偏好等问题。 https://lulietlyan.github.io/topics/ai-agent/evaluation/capability-evaluation https://lulietlyan.github.io/topics/ai-agent/evaluation/capability-evaluation Sat, 12 Sep 2026 00:00:00 GMT 按能力定义、数据集构建和评估方法梳理专项评估方案。 https://lulietlyan.github.io/topics/ai-agent/evaluation/agent-versus-llm-evaluation https://lulietlyan.github.io/topics/ai-agent/evaluation/agent-versus-llm-evaluation Sat, 12 Sep 2026 00:00:00 GMT 比较单轮文本质量与多步环境交互的评估维度及复现难度。 https://lulietlyan.github.io/topics/ai-agent/evaluation/agent-benchmarks https://lulietlyan.github.io/topics/ai-agent/evaluation/agent-benchmarks Sat, 12 Sep 2026 00:00:00 GMT 整理 Agent 基准中的任务、交互环境与能力覆盖方式。 https://lulietlyan.github.io/topics/ai-agent/evaluation/agent-process-metrics https://lulietlyan.github.io/topics/ai-agent/evaluation/agent-process-metrics Sat, 12 Sep 2026 00:00:00 GMT 介绍执行效率、成本、鲁棒性与其他过程指标的评估价值。 https://lulietlyan.github.io/topics/ai-agent/evaluation/red-teaming https://lulietlyan.github.io/topics/ai-agent/evaluation/red-teaming Sat, 12 Sep 2026 00:00:00 GMT 整理对抗测试的目标、攻击角度以及发现安全问题和偏见的作用。 https://lulietlyan.github.io/topics/ai-agent/evaluation/human-evaluation https://lulietlyan.github.io/topics/ai-agent/evaluation/human-evaluation Sat, 12 Sep 2026 00:00:00 GMT 介绍评分准则、标注流程、一致性检查和质量控制方法。 https://lulietlyan.github.io/topics/ai-agent/evaluation/production-monitoring https://lulietlyan.github.io/topics/ai-agent/evaluation/production-monitoring Sat, 12 Sep 2026 00:00:00 GMT 从日志、质量监控、人工分析到反馈迭代构建持续评估流程。